Physics-Informed Equipment Health Monitoring: Catching Degradation Before Failure

Rotating equipment—compressors, pumps, turbines—drives most of the operating cost and most of the unplanned downtime in Oil & Gas facilities. The standard monitoring approach hasn't changed much in decades: set a vibration threshold, set a temperature threshold, alarm when either is exceeded. This catches equipment that has already failed or is about to. It rarely catches equipment that is degrading.

Physics-informed health monitoring works differently. Instead of asking "is this reading outside its normal range," it asks "is this equipment converting energy the way its design physics says it should." That question surfaces degradation weeks or months before a threshold alarm would.

The Limits of Threshold-Based Monitoring

A vibration threshold alarm is a lagging indicator by construction. Bearing wear, misalignment, and impeller fouling all increase vibration amplitude, but only after the underlying mechanical or thermodynamic problem has progressed far enough to produce a detectable signature at the sensor. By the time the alarm trips, the failure mode is often already in its acceleration phase.

Statistical anomaly detection—flagging readings that deviate from historical norms—improves on fixed thresholds but has its own blind spot. It compares today's reading to yesterday's distribution, not to what the equipment's governing physics says the reading should be. A compressor that gradually loses efficiency across a six-month period may never look statistically anomalous, because the drift is slow enough that the "normal" distribution just quietly shifts along with it.

What Physics Brings to Health Monitoring

Every piece of rotating equipment has performance relationships that are supposed to hold regardless of how the machine is aging: conservation of energy, the pump or compressor's design performance curve, and (for compressors) the thermodynamic efficiency implied by the compression process itself.

These relationships give you an independent reference that doesn't drift with the equipment. When a machine's actual performance departs from what its physics predicts, that gap is the health signal—and it is measurable long before it manifests as vibration or temperature outside normal bounds.

Compressors: Isentropic Efficiency as the Physics Anchor

For a centrifugal or reciprocating compressor, the thermodynamics of compression provide a natural health metric: isentropic efficiency, the ratio of the ideal (reversible, adiabatic) work required to compress the gas to the actual work the machine is consuming for the same pressure ratio.

Isentropic efficiency is calculable directly from routinely available field measurements—suction and discharge pressure, suction and discharge temperature, gas composition, and driver power or flow. It is not a proxy; it is derived from the first and second laws of thermodynamics applied to the actual process gas.

A new or recently overhauled compressor operates close to its design efficiency. As internal clearances open up from wear, as seals degrade, or as fouling builds on impeller surfaces, the actual work required to hit a given pressure ratio increases relative to the ideal work—efficiency falls. This decline is typically gradual and monotonic, which makes it a strong candidate for trend-based prognosis rather than just point-in-time alarming: a compressor losing half a percentage point of isentropic efficiency per month is telling you something a vibration spectrum won't show for a long time.

Pumps: Affinity Laws and Head-Flow Curves

Centrifugal pumps offer a similar physics anchor: the design head-flow curve, together with the affinity laws relating flow, head, and power to impeller speed. At any operating speed, a healthy pump should produce a head that falls on (or very close to) its design curve for the corresponding flow rate.

Impeller wear, internal recirculation from worn wear rings, and cavitation damage all show up as a downward shift of the actual head-flow point relative to the design curve—before they show up as audible cavitation noise or a bearing vibration alarm. Because the head-flow relationship is a two-dimensional physics constraint rather than a single threshold, it's also more diagnostic: a curve shift concentrated at low flow points toward recirculation, while a uniform downward shift across the flow range points toward impeller wear or internal clearance growth.

Building the Health Score

A physics-informed health score combines the deviation from the governing physics relationship with an estimate of measurement uncertainty, so that a health indicator reflects genuine equipment drift rather than sensor noise or transient operating conditions.

The general structure:

1. Baseline the physics model. Establish the as-commissioned (or as-overhauled) performance curve or efficiency relationship for the specific asset, using either OEM data or early-life field data.

2. Compute the physics-predicted performance continuously. For every operating point, calculate what the governing equations say the machine's performance should be at that flow, pressure ratio, or speed.

3. Compare to actual performance. The residual between predicted and actual is the raw health signal.

4. Filter for operating regime and uncertainty. Not every residual reflects degradation—normalize for known effects like gas composition changes or ambient temperature, and weight by measurement confidence.

5. Trend, don't just threshold. A single bad reading is noise. A residual that grows steadily over weeks is a trend, and trends support a remaining-useful-life estimate rather than a binary alarm.

From Detection to Diagnosis

The physics-based residual doesn't just say something is wrong—the pattern of the residual is diagnostic. A compressor efficiency loss concentrated at high pressure ratios points toward valve or seal leakage; a loss that's uniform across the operating envelope points toward fouling. A pump head deficit at shutoff versus one that widens with flow implicates different wear mechanisms. This diagnostic specificity is what lets a maintenance planner schedule the right intervention, not just any intervention, during the next planned outage.

Deployment Considerations

This approach requires accurate baseline physics for each asset class and access to the process measurements the physics needs—which for compressors and pumps are typically already collected for other operational purposes. It does not require new instrumentation in most facilities, and it does not require months of failure-history data the way a purely data-driven predictive maintenance model does, since the physics constraint is valid from day one rather than learned from past failures.

The practical benefit is lead time. A physics-informed health score that has been trending down for six weeks gives a maintenance planner time to schedule a parts order and an outage window. A vibration alarm that trips gives them an unplanned trip instead.

---

WellBeyond.ai builds physics-informed health monitoring for rotating equipment across upstream, midstream, and downstream operations. Talk to us about the assets you're trying to keep running.